Valerio Scorretti's Work | ContraWork by Valerio Scorretti
Valerio Scorretti

Valerio Scorretti

AI Solutions Architect & Agentic Workflow Engineer

New to Contra

Valerio is ready for their next project!

Followed by Sebastien G and Eman K

Italy

WorkPostsServicesAbout

Italy

AEGIS: Enterprise PII Masking & Secure LLM Gateway (Active Beta / WIP) Architecting a model-agnostic, privacy-first AI interface designed for strict data compliance. AEGIS selectively intercepts and masks PII (Personally Identifiable Information) before the payload reaches external LLMs. Core Architecture & Features: Granular PII Engine: Automated selective pseudonymization with manual token override (Brush tool for custom inclusion/exclusion). Dual RAG System: Ephemeral Session-RAG for active context, and persistent Knowledge Base RAG for long-term memory retrieval. Stateful Chat Retrieval: Hash-based code injection to instantly load and vectorize past conversations into the active prompt box. Model-Agnostic Privacy: Seamlessly switch between top-tier LLMs while maintaining the local data-masking shield. Analytics & Benchmarking: Built-in model benchmarking, real-time token consumption tracking, and modular feature-toggling via admin panels. Native Multilingual: Engineered for English, Spanish, Italian, and Catalan. Current Status: WIP / Active Beta. Currently debugging state management and stabilizing the RAG pipelines before production release.
0
13
Vertex BCN: Real-Time Autonomous Urban Agent Architected and deployed a production-grade, multilingual AI assistant for urban navigation. This is a complex Agentic Workflow, not a basic LLM wrapper. Core Architecture: Real-time API integrations for dynamic transit and scheduling data. Custom RAG pipelines for contextual location discovery. Low-latency LLM orchestration handling parallel multilingual queries. Engineered for fault tolerance, scalability, and zero downtime. I build the backend infrastructure that ensures reliable, enterprise-level data delivery.
0
17
Cover image for End-to-End Agentic Workflows (n8n) &
End-to-End Agentic Workflows (n8n) & AI-Blender 3D Pipelines
0
70
Cover image for Stop treating LLMs as your
Stop treating LLMs as your software architects. They are just the construction workers. Many startups burn precious runway on raw API calls without system design. If your AI infrastructure lacks strict data isolation, sub-second latency (WebRTC), and localized business logic, you are building an expensive toy, not an asset. To slash OpEx, you must architect the system first, then let the LLMs do the labor. What is your biggest bottleneck when deploying AI to production? Let's discuss.
0
71